Announcement

Collapse
No announcement yet.

NVIDIA Linux Vulkan Performance vs. RADV / AMDGPU-PRO

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #21
    I assume if Michael has Fury X that would with Vulkan eat all greens there

    Comment


    • #22
      Hey Michael. Although I have a desktop with an old Nvidia card, I have 3 laptops running various AMD APUs. One with a Mullins APU, one with a Kaveri and one with a Carrizo.

      Could you do a similar test like the above on various AMD APUs since that seems to be a big focus with AMD ? Forget about Nivida of course.....just looking at the performance differences between different generations of AMD APUs

      Also....since AMD just released both the laptop and desktop versions of Bristol Ridge ( the Carrizo refresh ) could you do a benchmark again showing the differences in performance in DDR 3 and DDR 4 speeds like DDR 3 1600 vs DDR 3 2400 or DDR 4 2400 ?

      Comment


      • #23
        On APUs situation can be entirely different, for example see this sort of recent on A10-7850K where blob is actually 16% faster then mesa:

        Comment


        • #24
          But i would like to see amdgpu and amdgpu-pro on APUs too, this was radeon vs fglrx... so either blob regressed somewhere or mesa filmsy memory managment strikes back on APUs

          Comment


          • #25
            Originally posted by dungeon View Post
            But i would like to see amdgpu and amdgpu-pro on APUs too, this was radeon vs fglrx... so either blob regressed somewhere or mesa filmsy memory managment strikes back on APUs
            That link also shows the test is with Mesa 12 (versus current git, which contains a lot of optimizations in radeonsi driver) and is only providing GL 4.1 support, which in turn likely means they are running with LLVM 3.7, which again is missing tons of optimization work done since then including the modular shaders/partial caching support added in 3.8.
            Last edited by smitty3268; 09 October 2016, 05:21 PM.

            Comment


            • #26
              It is also fglrx versus current amdgpu-pro, that contain probably even more optimizations nearly a year more

              Comment


              • #27
                I mean if we compare things be real, fglrx is very old and even latest amdgpu-pro release is more then 2 months old... so what we compare there, latest git of all against something that old

                Comment


                • #28
                  BTW; Windows driver as of 16.9.1 i think switched to new 16.40 serie... maybe we should wait for that and then compare it

                  Comment


                  • #29
                    Originally posted by dungeon View Post
                    It is also fglrx versus current amdgpu-pro, that contain probably even more optimizations nearly a year more
                    Perhaps, or perhaps not. Mesa is a much faster moving target since it was so much slower to begin with.

                    It's quite believable that the proprietary drivers would handle the lesser memory bandwidth better than the Mesa drivers do, I'm just saying that the benchmark you linked to doesn't really justify that IMO because it's not very relevant to the current drivers or the current tests on other hardware we've seen Michael run.
                    Last edited by smitty3268; 09 October 2016, 06:15 PM.

                    Comment


                    • #30
                      There as i see is 4.5.13399 which is Catalyst 15.9 so more then year old something, against still 12 release of mesa and still default radeon driver for CIK... so i see that as all correct not pushed over comparison

                      Comment

                      Working...
                      X